Business Continuity is a critical business factor in general for SAP systems. Business Continuity includes the ability to avert unplanned outages and the ability to avoid or minimize planned downtime.
IBM┬« Db2┬« with pureScale┬«, a proven technology, was designed to deliver high availability and disaster recovery from the start. IBM┬« Db2┬« pureScale┬« is continuously enhanced, and its recent iteration supports the avoidance of planned downtimes and delivers a rich set of capabilities to help ensure Business Continuity.
Architectural overview of Db2 pureScale
The Db2 pureScale feature is an extension to the existing Db2 product. You can use Db2 pureScale to create a database cluster by applying the shared disk approach.
Figure 1. Overview of Db2 pureScale architecture
A node in a Db2 pureScale cluster is called a member. Each member runs a Db2 instance that uses the Db2 server software and has access to the complete database.
Db2 pureScale uses the cluster caching facility (also known as CF) to assist data sharing between members and to address the problems of concurrency control and cache coherency. The CF typically runs on a dedicated physical or logical host and is connected to all members of the cluster that uses a high-speed interconnect. This high-speed interconnect is realized using an InfiniBand fabric or RDMA over Converged Ethernet (RoCE) network. Alternatively, the interconnect can be realized with a 10-Gigabit TCP/IP Ethernet (10GE) network infrastructure.
The disk subsystem is accessed by all members of the pureScale cluster, and must, therefore, allow for concurrent read and write access to maintain the integrity of the data. A clustered file system (also known as shared disk file system) is used to help achieve this. Db2 pureScale uses IBM Spectrum ScaleÔäó, also known as IBM┬« General Parallel File System (GPFSÔäó). GPFS is a high-performance, scalable file management solution that provides fast and reliable access to a common set of files. Management and installation of GPFS is transparent to the user and embedded in Db2 pureScale.
Another component that is integrated into Db2 pureScale is the cluster manager. The cluster manager checks the availability of all cluster components. If it detects the failure of a component, it automatically performs all necessary actions so that the impact on the overall system is as minimal as possible.
The combination of the components described leads to an integrated database cluster that supports Business Continuity with high availability, minimized planned downtime, and disaster recovery functions.
SAP enhancements for Db2 pureScale
To achieve the business continuity targets, not only does the database need to be considered, but the application requires optimization and enhanced administration capabilities to get the most out of the solution. Therefore, SAP developed and integrated some features to optimize the usage of Db2 pureScale with SAP applications. For details and recent updates, check out the SAP on IBM Db2 pureScale landing page within the SAP Community Wiki.
Client affinity concept
With Db2 pureScale, several database members serve the application workload. The best resource usage and throughput is achieved by a uniform use of the members. To realize a similar workload on each member:
- On the SAP application level, use logon groups to direct SAP users with related tasks to a specified set of SAP application servers.
- On the database level, you can assign every member to one or more SAP application servers by using a simple round-robin assignment of application servers to database members or a custom assignment of members to one or more application servers.
The custom assignment also allows various distinctive designs for the pureScale usage, including a particular member serving for failover purposes only or solely used for administrative purposes, such as backups. Together with the application level workload balancing and operation modes, a wide range of optimized assignments are possible.
Figure 2. Db2 pureScale Setup with dedicated spare member
Partitioning of critical tables
SAP applications based on ABAP NetWeaver typically have a number of tables with high concurrent workload. SAP supports the partitioning of database tables and indexes by using the member, the application server, or other generated values as partitioning criteria. With this, you can optimize the concurrency and ensure the scalability.
Db2 pureScale monitoring functions via the SAP DBA cockpit.
The SAP DBA Cockpit was enhanced with specific functions to enable monitoring Db2 pureScale members and CFs. These Db2 pureScale-specific enrichments provide:
- An overview of the cluster topology and cluster alerts
- An overview of the Db2 member performance in the Db2 pureScale cluster
- Information about the group buffer pool and local buffer pool hit ratio
- The content of the (global) package cache
- Information about the memory configuration and consumption of the cluster caching facilities (CFs)
- Information about page contention in the Db2 pureScale cluster
- Information about connected clients and the client configuration.
Disaster recovery options with pureScale
In addition to the core capabilities included in the basic pureScale architecture, there are options available to enhance the database availability. Excluding the classical outages due to local failures of hardware or software components and the required maintenance efforts that require downtime, there might be actions that are required to protect against complete computing center outages due to external distractions. One option to contain those issues is to replicate the IT environment to a secondary site, which is located a certain distance away. Db2 pureScale supports disaster recovery (DR) by using replication methods on disk storage level or the replication capabilities of IBM spectrum scale (GPFS). Two additional capabilities allow for a more sophisticated implementation. Db2 pureScale supports an implementation setup that is called geographically dispersed cluster (GPPC). In a GDPC setup, the cluster is stretched across two sites with local disk access on each site, which is synchronized by using the GPFS functions. From the application point of view, the cluster still acts as one cluster.
Another approach is to use the well-known capability of Db2 called HADR. In this scenario, a pureScale cluster is duplicated, and data is sent to the secondary site that uses the HADR functions.
Figure 3. Db2 pureScale setup with HADR
Db2 pureScale deployment options
Db2 pureScale with the minimal configuration for high availability requires two physical hosts and is supported by SAP applications that run on IBM Power Servers with AIX and x86 based servers running under Linux. Db2 pureScale supports multiple virtualization offerings like PowerVM on IBM Servers, KVM, and VMware on x86 based servers.
Especially with the option to use standard 10 GB Ethernet as the cluster interconnect network, this allows for a flexible and low-cost entry point into the world of Db2 pureScale. The 10 GB Ethernet support allows you to set up development or quality assurance systems and production systems if performance requirements are compatible with the socket data transport.
For more challenging workloads, you can deploy the cluster interconnect network with remote direct memory access (RDMA) capable adapters that use InfiniBand or RDMA over Converged Ethernet (RoCE) Adapters.
Figure 4. Db2 pureScale deployment examples
Various deployment options are available for Db2 pureScale. It is possible to have multiple pureScale nodes on one physical server or on dedicated physical machines (for the different scenarios see Figure 4).
It is also possible to equip each member and CF with multiple network interconnect adapters and connect each adapter to a dedicated network switch. There is no single point of failure within the cluster with this setup.
Online software update
Db2 pureScale enables the administrator to upgrade the operating system without a system outage. The member running on the physical or virtual machine pauses, and all connections are transferred to the remaining members. Operating system maintenance is next, then the member on the updated operating system is integrated again.
Similar functionality called rolling Fick Pack RW: Is this really called Fick Pack? updates is available for Db2 Fix Packs. During the online installation of a new Fix Pack, the member or CF is updated to the new software level but still operates with the previous level. After all nodes in the cluster are updated, the new software level activates, and the cluster is transparently updated to the new Db2 Fix Pack Level.
You can add new members to an instance while it is online without an impact to the workloads running on the existing members. No additional database tuning or data redistribution is required. Db2 members are individually tuned by the Db2 self-tuning memory manager, as soon as they are part of the cluster. Depending on the client connectivity setup, new database members are automatically used by the new SAP application servers or after the new client setup is reloaded via the SAP DBA Cockpit. There is no need to stop or restart the SAP application.
Dropping a member still requires the cluster to be offline. However, for temporarily decreasing workloads, one or more members can be brought offline, and their resources can be freed up.
Db2 pureScale conclusion
The pureScale option is included (no additional license costs) Db2 edition that is certified for SAP Software (Db2 Advanced Enterprise Server Edition).
The pureScale feature simplifies the deployment of a distributed database environment. It is easy to deploy pureScale during an SAP implementation because it is bundled and deeply integrated within the standard SAP installation process and administration tools like the SAP DBA Cockpit.
With Db2 pureScale, you can easily set up a high availability and disaster recovery (HADR) scenarios as part of data center readiness. Db2 pureScale also provides HADR capabilities with multiple options to select RTO (recovery time objective) and RPO (recovery point objective) timelines based on planned and unplanned downtime for your SAP IT environment.